Drone detection with micro-doppler analysis
Drone detection now plays an instrumental role in security concepts. The use of micro-Doppler analysis to determine additional characteristics of the flying objects, such as the object class, size and weight, facilitates an estimation of the level of risk caused by the drone.
Compact radar devices can be used to monitor local airspace so as to protect objects or events from threats posed by drones. The strength of the signal received allows conclusions to be drawn with regard to the approximate size of a detected object, but further analysis is, however, not possible. Further information relating, for example, to the type of the detected object (bird, fixed-wing airplane, helicopter, quadcopter, octocopter), the size of the object and the possible load capacity is desirable. A special radar mode and micro-Doppler analysis is necessary to draw conclusions relating to the number of propellers and their rotational speed.
Construction of a micro-Doppler demonstrator
The demonstrator DroMiAn (Drone detection with micro-Doppler analysis), which, in addition to the normal radar hardware for object detection, also has an additional channel for micro-Doppler measurement, was constructed to research the micro-Doppler signatures. This demonstrator operates at a fixed frequency of 94 GHz. In a first step, the demonstrator was used to measure rotating objects in the measurement chamber. Free-field measurements with a flying drone were carried out at a later stage.
Measurements in the measurement chamber
Rotating bodies of the same length, including a tube and different types of propellers, were used in the first test measurements with the demonstrator. These were measured at varying rotational speeds and inclinations. The frequency and speed of the rotating objects can be clearly determined from the recorded data. These two values can then be used to calculate the size of the object.
The first free-field test measurements involved the measurement of a quadcopter in different flight scenarios, i.e. hanging motionless in the air, flying slowly from left to right, flying slowly towards, away from and over the radar device. In addition, comparative data of individual pedestrians and groups of pedestrians was recorded to create false targets for the evaluation algorithms.
Compared to the measurements in the measurement chamber, the evaluation of the real free-field measurements proved to be much more difficult as, instead of measuring the signature of a single propeller, it was now necessary to measure a sum signal made up of the reflections from several propellers and the body of the drone.
The measurement data was evaluated in several steps. The first step involved creating a spectrogram from the complex radar data. This is made up of a series of Fourier transformations which are specially adapted for the respective conditions so as to achieve the best possible results. Before processing continues, this spectrogram is optimized through the suppression of fixed targets and the removal of noise components. In a further processing step, features are extracted by means of singular value decomposition. This technique provides good results, particularly in the case of the individual pedestrians or the pedestrian groups. In the case of the measured drone, the features were extracted from a cepstrogram. This is a technique originally used for frequency analysis in the area of sound technology. When used in conjunction with radar data, it provides good results when determining the rotational speed of the rotors. The individual rotor speeds can subsequently be read from the cepstrogram or the cepstrum, a cross-section of the cepstrogram.
Target objects are classified by comparing the measured data with the database of known objects. As a sufficiently large database with measured objects was not available for the tested frequency range, synthetically generated data from drones, birds and other objects with complex dynamics was used. In the first classification tests, a correct classification rate of 86% was achieved for the measured drones and the classifications for individual pedestrians and pedestrian groups were 100% correct.
In the meantime, new data sets for drones of different types have been recorded with the further developed radar demonstrator. This data will be analyzed with new signal processing and target classification approaches.